[GE users] H_vmem question
reuti at staff.uni-marburg.de
Mon Dec 22 22:57:12 GMT 2008
Am 22.12.2008 um 23:36 schrieb udo:
> A simple question:
> Objection: I want to avoid to use swap space on compute nodes.
> I have 8 core Xeon machines with 8 GB of RAM on each and 4 GB of
> swap i.e.
> total 12 GB of virtual memory.
> So, what should I put in "Limits"
> 1) h_vmem = 8GB meaning total on a node or
> 2) h_vmem = 1GB meaning each process should take not more than 1GB
> i.e. 1GB
> x 8 (core) = 8GB per node.
> Thank you very much for any valuable answer.
a) make h_vmem consumable in `qconf -mc`
b) give it a proper default value in `qconf -mc` which each process
will get, unless the user requets something else, e.g. 1GB
c) for each exec host `qconf -me <nodename>` define "complex_values
h_vmem=8GB", i.e. the physical memory: so max 8GB could be requested
in total on each node for all jobs running there at the same time
d) optional: in the queue definition `qconf -mq <qname>` you can
define an upper limit for each process (in case you need a lower
limit than the defined 8GB in the exechost definition), e.g
h_vmem=6GB to avoid that already one job would consume all the memory
users can request up to: 6GB
total available per node: 8GB
Be aware that h_vmem is an enforced limit and jobs will get killed
when they exceed this limit. You might also want to switch on
resource reservation and submit jobs with "qsub -R y ...".
To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
More information about the gridengine-users